Hybrid new energy station small interference stability margin prediction method and device

Abstract

The invention discloses a hybrid new energy station small interference stability margin prediction method and device, and the method comprises the steps: obtaining system operation state data, and carrying out the dimension reduction processing of the system operation state data; constructing an edge convolution module based on an edge graph convolution network; the edge convolution module is embedded into a Transform model based on a topology enhanced self-attention mechanism, and an optimized Transform model is obtained; and inputting the system operation state data subjected to dimension reduction processing into the optimized Transform model, and performing prediction through the optimized Transform model to obtain a hybrid new energy station grid-connected stability margin prediction value, thereby realizing hybrid new energy station small-interference stability margin prediction. According to the method, the small-interference stability margin prediction precision of the hybrid new energy station can be improved, so that the requirement of safe and stable operation of a power grid can be met.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and device for predicting small disturbance stability margin of a hybrid new energy station. Background Art

[0002] In the process of building new power systems, with the access of a high proportion of new energy and power electronic equipment, the power system has obviously shown a trend of reduced inertia and weakened system strength, and the stability problem has become more serious. Therefore, it is of great practical significance to study and analyze the small disturbance stability and stability margin of the hybrid renewable energy station (HRPS) under weak or even extremely weak power grid conditions.

[0003] The configuration of grid-forming (GFM) / grid-following (GFL) type control generators and energy storage devices in HRPS (capacity ratio and location distribution in the station) and the grid strength (short-circuit ratio) of the external power system will affect the small disturbance stability of the grid-connected system; further, the topology and power flow changes of the collector network in HRPS also affect the small disturbance stability of the grid-connected system. The factors such as the time-varying strength of the external grid caused by the frequent changes in the operation mode of the new power system have brought difficulties to the online evaluation of the small disturbance stability of the grid-connected system in a model-driven manner. It is necessary to use artificial intelligence technologies such as deep learning to realize the online evaluation of the small disturbance stability margin of the grid-connected system in a data-driven manner. The existing data-driven methods do not take into account the changes in the collector network topology and power flow in HRPS, which affects the prediction accuracy of the small disturbance stability margin. Summary of the invention

[0004] The present invention aims to solve the technical problems in the related art at least to a certain extent. To this end, the first purpose of the present invention is to provide a method for predicting the small disturbance stability margin of a hybrid new energy station, which can adapt to the changes in the topology and collection network flow in HRPS, and output the predicted values ​​of the small disturbance stability margin of the low frequency band and the sub / super synchronous frequency band online according to the external power grid strength, the capacity ratio and location distribution of the grid-building / grid-following generator sets in HRPS, the control type of the energy storage device, the topological structure and other information. It not only has higher prediction accuracy, but also the prediction results tend to be conservative, which can meet the requirements of safe and stable operation of the power grid.

[0005] The second object of the present invention is to provide a hybrid new energy station small disturbance stability margin prediction device.

[0006] A third object of the present invention is to provide a computer-readable storage medium.

[0007] A fourth object of the present invention is to provide an electronic device.

[0008] To achieve the above object, the present invention is implemented through the following technical solutions:

[0009] A method for predicting small disturbance stability margin of a hybrid new energy station, comprising:

[0010] Obtain system operation status data and perform dimensionality reduction processing on the system operation status data;

[0011] Construct an edge convolution module based on edge graph convolutional network;

[0012] The edge convolution module is embedded into the Transformer model based on the topology-enhanced self-attention mechanism to obtain the optimized Transformer model;

[0013] The system operation status data after dimensionality reduction processing is input into the optimized Transformer model. The predicted value of the grid-connected stability margin of the hybrid new energy station is predicted by the optimized Transformer model, thereby realizing the small disturbance stability margin prediction of the hybrid new energy station.

[0014] Preferably, the system operation status data includes the strength of the external power grid, the control type, capacity ratio and location distribution of the generator sets in the HRPS, and the control type of the energy storage device at the grid connection point.

[0015] Preferably, the system operation status data is subjected to dimensionality reduction processing based on a T-distributed random neighbor embedding algorithm.

[0016] Preferably, constructing an edge convolution module based on an edge graph convolutional network includes:

[0017] Obtain HRPS topology data and collector network flow data;

[0018] The edge weight is calculated according to the branch flow information in the collector network flow data to assist the edge convolution module to be constructed to learn the HRPS topology data;

[0019] Convolution operation is performed on the features of nodes and edges in the HRPS topology data to update the nodes, wherein the nodes are power generation units in the HRPS topology structure and the edges are lines between the power generation units;

[0020] The ReLU activation function is used to perform nonlinear transformation processing on the update node, and the processed updated node features are used as input to perform the convolution operation of the next layer. After multiple layers of convolution operations, the edge convolution module is formed.

[0021] Preferably, power flow calculation is performed based on HRPS topology data based on a holomorphic embedding method to obtain the power collection network power flow data.

[0022] Preferably, the optimized Transformer model weights the attention score through the embedded edge convolution module when performing self-attention calculation.

[0023] To achieve the above-mentioned object, the second aspect of the present invention provides a hybrid new energy station small disturbance stability margin prediction device, comprising:

[0024] A data processing module is used to obtain system operation status data and perform dimensionality reduction processing on the system operation status data;

[0025] A building module for building edge convolution modules based on edge graph convolutional networks;

[0026] The embedding module is used to embed the edge convolution module into the Transformer model based on the topological enhanced self-attention mechanism to obtain the optimized Transformer model;

[0027] The prediction module is used to input the system operation status data after dimensionality reduction processing into the optimized Transformer model, and obtain the predicted value of the grid-connected stability margin of the hybrid new energy station through the optimized Transformer model, thereby realizing the small disturbance stability margin prediction of the hybrid new energy station.

[0028] Preferably, when constructing an edge convolution module based on an edge graph convolutional network, the construction module is specifically used to:

[0029] Obtain HRPS topology data and collector network flow data;

[0030] The edge weight is calculated according to the branch flow information in the collector network flow data to assist the edge convolution module to be constructed to learn the HRPS topology data;

[0031] Convolution operation is performed on the features of nodes and edges in the HRPS topology data to update the nodes, wherein the nodes are power generation units in the HRPS topology structure and the edges are lines between the power generation units;

[0032] The ReLU activation function is used to perform nonlinear transformation on the update node, and the processed updated node features are used as input for the next layer of convolution operation. After multiple layers of convolution operations, an edge convolution module is formed.

[0033] To achieve the above-mentioned purpose, the third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned method for predicting small disturbance stability margin of a hybrid new energy station is implemented.

[0034] To achieve the above-mentioned purpose, the fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for predicting small disturbance stability margin of hybrid new energy station is implemented.

[0035] The present invention has at least the following technical effects:

[0036] On the one hand, in order to solve the problem of too high dimensionality of the model input feature data, the present invention adopts the T-distribution random neighbor embedding algorithm to perform data dimensionality reduction processing, which reduces the computational complexity and improves the quality of the input feature data. On the other hand, the present invention constructs voltage and power holomorphic functions by embedding complex variables in the collector network flow equation, and uses the holomorphic embedding method to realize the rapid calculation of the flow; on this basis, the edge graph convolutional network is introduced to construct the edge convolution module, which not only considers the topological connection in the HRPS, but also takes the flow information calculated by the holomorphic embedding method into consideration as the edge feature of the edge convolution module. Compared with the traditional graph neural network, the edge convolution module based on the edge graph convolutional network not only realizes the perception of the HRPS topology, adapts to the topological changes within the station, and improves the topological generalization ability of the model, but also can obtain the flow characteristics of the collector network, providing richer feature input for the subsequent small disturbance stability margin prediction. On the basis of the first two, the present invention aims at the self-attention mechanism in the traditional Transformer model, which quantifies the global correlation between input data through dot product calculation and ignores the topological correlation problem of input features. The edge convolution module based on the edge graph convolution network is embedded into the self-attention mechanism of the Transformer model. The nodes and edge information in the graph are introduced into the self-attention calculation by constructing an edge convolution module and improving the attention matrix, so that the optimized Transformer model can better fit the nonlinear relationship between the edge graph characteristics of HRPS and the operating status characteristics of the grid-connected system and the small disturbance stability margin, thereby improving the prediction accuracy of the Edge-GCN-Transformer deep learning model.

[0037] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The present invention is a flowchart of a method for predicting a small disturbance stability margin of a hybrid new energy station according to an embodiment of the present invention.

[0039] Figure 2 The present invention is a general flow chart of a method for predicting a small disturbance stability margin of a hybrid new energy station according to an embodiment of the present invention.

[0040] Figure 3This is a structural diagram of a HRPS network / grid-following device including an energy storage device according to an embodiment of the present invention.

[0041] Figure 4 It is a structural block diagram of a hybrid new energy station small disturbance stability margin prediction device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present embodiment is described in detail below, and examples of the embodiment are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0043] The following describes the method and device for predicting the small disturbance stability margin of a hybrid new energy station of this embodiment with reference to the accompanying drawings.

[0044] Figure 1 The present invention is a flowchart of a method for predicting a small disturbance stability margin of a hybrid new energy station according to an embodiment of the present invention. Figure 2 FIG. 1 is an overall flow chart of a method for predicting a small disturbance stability margin of a hybrid new energy station according to an embodiment of the present invention. Figure 1 and Figure 2 As shown, the method includes:

[0045] Step S101: Acquire system operation status data, and perform dimensionality reduction processing on the system operation status data.

[0046] Among them, the system operation status data includes the external grid strength, the control type of the generator sets in the HRPS, the capacity ratio and location distribution, and the control type of the energy storage device at the grid connection point.

[0047] In this embodiment, the system operation status data is subjected to dimensionality reduction processing based on the T-distribution random neighbor embedding algorithm.

[0048] Step S102: construct an edge convolution module based on an edge graph convolutional network.

[0049] Among them, constructing an edge convolution module based on an edge graph convolutional network includes: obtaining HRPS topological data and collector network flow data, wherein the flow calculation is performed based on the HRPS topological data based on the holomorphic embedding method to obtain the collector network flow data; calculating the edge weight according to the branch flow information in the collector network flow data to assist the edge convolution module to be constructed to learn the HRPS topological data; performing convolution operations on the features of the nodes and the features of the edges in the HRPS topological data to update the nodes, wherein the nodes are the power generation units in the HRPS topological structure, and the edges are the lines between the power generation units; using the ReLU (linear rectification function) activation function to perform nonlinear transformation processing on the updated nodes, and using the processed updated node features as input to perform convolution operations on the next layer, and after multiple convolution layer operations, an edge convolution module is formed.

[0050] Step S103: embed the edge convolution module into the Transformer model based on the topology enhanced self-attention mechanism to obtain an optimized Transformer model.

[0051] Step S104: input the system operation status data after dimensionality reduction processing into the optimized Transformer model, and obtain the predicted value of the grid-connected stability margin of the hybrid new energy station through the optimized Transformer model to realize the small disturbance stability margin prediction of the hybrid new energy station.

[0052] Among them, the optimized Transformer (a machine learning model for prediction) model weights the attention score through the embedded edge convolution module when performing self-attention calculation.

[0053] In order to clearly describe the small disturbance stability margin prediction method for a hybrid new energy station of the present embodiment, the small disturbance stability margin prediction method for a hybrid new energy station is described in detail below.

[0054] A small disturbance stability margin prediction method for a hybrid network / following new energy station based on an edge graph convolutional network and a Transformer fusion model includes the following steps:

[0055] Step 1: Based on the T-distributed Stochastic Neighbor Embedding (TSNE) algorithm, the high-dimensional feature input data composed of system operation status data including external grid strength (short-circuit ratio), control type (grid building / grid following) and capacity ratio and location distribution of generator sets in HRPS, and control type (grid building / grid following) of energy storage devices at grid connection points are reduced in dimensionality and then provided to the Transformer deep learning model to reduce computational complexity.

[0056] Step 2: Construct an edge convolution module based on the edge graph convolutional network (Edge-GCN), which constructs an edge weight matrix based on node features and edge features, and then updates the representation of nodes and edges through convolution operations. In this embodiment, the edge convolution module can combine the features of nodes and edges in each convolution layer to achieve perception of HRPS topology and improve the topological generalization ability of the deep learning model.

[0057] It should be noted that, in this embodiment, the HRPS power flow can be quickly calculated based on the collector network topology in the HRPS based on the Holomorphic Embedding Loadflow Method (HELM), and the calculated collector network power flow data can be used as the Edge-GCN edge feature, thereby providing richer feature input for small disturbance stability margin prediction.

[0058] Step 3: Embed the edge convolution module into the self-attention mechanism of the Transformer model to optimize the topological perception ability of the self-attention mechanism in the Transformer model, so that the Transformer model can better fit the edge graph characteristics of HRPS and the nonlinear relationship between the operating status characteristics of the grid-connected system and the small disturbance stability margin, thereby improving the prediction accuracy of the Edge-GCN-Transformer (deep learning model after fusion) model.

[0059] Optionally, the computational process of constructing an edge convolution module based on an edge graph convolutional network is as follows:

[0060] 1) Input data, including HRPS topology structure (i.e., HRPS topology data) and collector network flow data quickly calculated by the embedded HELM algorithm.

[0061] 2) The edge weights are calculated based on the branch flow information in the collector network flow data to assist the edge convolution module to better learn the topological structure and physical characteristics of the HRPS, as shown in formula (1):

[0062]

[0063] Where P ij , I ij , Z ij , ΔV ij are the power flow, current, impedance and voltage difference on branch ij respectively; P max , I max , ΔV maxare the maximum power flow, maximum current value and maximum voltage difference on branch ij respectively; α1, α2, α3, α4 are the first to fourth weight coefficients, indicating the influence of each factor on the edge weight; w ij is the edge weight of branch ij, i and j represent two different nodes, namely, power generation units.

[0064] 3) Update the node representation by convolving the node features with the edge features using formula (2).

[0065]

[0066] In the formula, represents the feature of node i in layer k, represents the feature of node i in the k+1 layer, σ(·) represents the Sigmoid activation function, Indicates the number of nodes, d i and d j is the degree of nodes i and j (i.e. the number of edges connected to these nodes), and L is the learnable weight matrix of the convolutional layer.

[0067] 4) Use the ReLU activation function to perform nonlinear transformation on the node update, and use the updated node features as input to perform the next layer of convolution operation. After two to three convolution operations, the edge convolution module based on the edge graph convolution network is constructed.

[0068] In step 3, the edge convolution module is embedded into the self-attention mechanism of the Transformer model to optimize the topological perception ability of the self-attention mechanism in the Transformer model, aiming to introduce the graph data represented by the edge graph convolution network into the Transformer model, so that the optimized Transformer model can better capture the topological structure correlation between the nodes in the data. When calculating the self-attention, the optimized Transformer model can introduce the edge convolution module to weight the attention score, and adjust the contribution of the topological information to the attention matrix in an adaptive manner, as shown in formula (3), so that the optimized Transformer model can effectively represent the topological structure dependency between the nodes in HRPS.

[0069]

[0070] Where A represents self-attention; softmax(·) represents the Softmax activation function; Q is the query matrix; E is the key matrix; d e is the dimension of the key; A adj is the adjacency matrix of the graph, which represents the connection relationship between nodes; w edgeis the weight of the edge; FC stands for the fully connected layer, according to A adj and w edge The weights added come from adaptive calculations.

[0071] Application examples:

[0072] In order to verify the feasibility and superiority of the present invention, Figure 3 The grid-building / grid-following HRPS with energy storage device shown in the figure generates a small disturbance stability assessment data set for the object. The internal generator sets of this HRPS are all direct-drive wind turbines, which are divided into two types: grid-building control and grid-following control according to the control mode of the converter. The hybrid new energy station adopts a chain structure with multiple feeders, including a total of 90 wind turbines, and the unit capacity is 2MW. There are five feeders in total, each of which connects 18 wind turbines. The wind turbines on the same feeder are connected through a bus line with a length of 0.43km. The inductance and resistance per unit length are 1.23mH / km and 0.17Ω / km respectively. A 20MW energy storage device is connected to the machine side at the PCC (common connection point).

[0073] The short-circuit ratio of the external power grid is set to 1 to 3 with an interval of 0.2, and there are 11 short-circuit ratio scenarios. Considering the two units at the head end of each feeder in HRPS (a total of 10 units), they can choose to follow the grid or build the grid. Each unit is a 0 / 1 variable. The energy storage device at the grid connection point can choose to follow the grid or build the grid. It is a 0 / 1 variable. Considering the topological form of interconnection between different feeder ends within HRPS, there are 10 topological types in total. Therefore, each short-circuit ratio scenario contains 20480 samples (2 10 ×2×10=20480), 11 short-circuit ratio scenarios with a total of 225280 samples, the data structure of the samples is shown in Table 1. In order to study the interaction between HRPS and the power grid, dq impedance modeling is used to calculate the characteristic value and damping ratio of the grid-connected system, and quantitatively evaluate the small disturbance stability and stability margin of the entire grid-connected system. For each short-circuit ratio scenario sample set, the training set and the test set are divided into 3:1, that is, samples are extracted from the sample set of each short-circuit ratio scenario in a ratio of 15360:5120 to form the training set and the test set.

[0074] The short-circuit ratio is set to 0.8 and 3.2 for a total of 2 scenarios, and the generalization test sample set is generated in the same way. There are 20480 samples in each short-circuit ratio scenario, and a total of 40960 samples in the two scenarios. The data structure of the samples is the same as that of the training set and the test set.

[0075] Table 1 Sample data structure table

[0076]

[0077] Regarding the criterion for the stability of small disturbances corresponding to the samples, the present invention discusses the stability of small disturbances in the low frequency band and the sub / super synchronous frequency band respectively. For the low frequency band, a damping ratio greater than or equal to 5% is strong damping, a damping ratio greater than or equal to 0 and less than 5% is weak damping, and a damping ratio less than 0 is instability; for the sub / super synchronous frequency band, a damping ratio greater than or equal to the dynamic critical damping ratio (set according to the needs of power grid safety and stability) is strong damping, a damping ratio greater than or equal to 0 and less than the dynamic critical damping ratio is weak damping, and a damping ratio less than 0 is instability. The sample set division is shown in Table 2.

[0078] Table 2 Sample set division table

[0079]

[0080] In order to quantitatively evaluate the effectiveness and superiority of the present invention, the small disturbance stability evaluation indexes of the present invention include the prediction optimism rate, the prediction optimism error mean, the prediction optimism error maximum, the prediction conservative rate, the prediction conservative error mean, the prediction conservative error maximum, the overall prediction error mean and the overall prediction error maximum, that is, the predicted stability margin prediction value can be evaluated by the above indicators. The definitions of each indicator are as follows:

[0081] 1) Forecast optimism rate (OPTI%), forecast optimistic error mean (EA_OPTI), forecast optimistic error maximum (EM_OPTI):

[0082] If the predicted value of the stability margin is greater than the actual value of the stability margin, the sample prediction result is defined as optimistic, so the following evaluation indicators are defined:

[0083] Forecast Optimism Rate (OPTI%):

[0084]

[0085] Forecast optimistic error mean (EA_OPTI):

[0086]

[0087] Maximum optimistic error of prediction (EM_OPTI):

[0088] EM_OPTI=Max(R_Pred_opti[n]-R_opti[n]), n∈[1, m] (6)

[0089] Where num_opti is the number of optimistic samples for stability margin prediction; m is the number of test samples; R_Pred_opti[n] and R_opti[n] are the predicted value and true value of the optimistic stability margin of the nth test sample, respectively.

[0090] 2) Forecast conservative rate (CONSERV%), forecast conservative error mean (EA_CONSERV), forecast conservative error maximum (EM_CONSERV):

[0091] If the predicted value of the stability margin is less than or equal to the actual value of the stability margin, the sample prediction result is defined as conservative. The higher the prediction conservative rate (CONSERV%), the more the stability margin prediction result meets the requirements of the power grid for safe and stable operation. Therefore, the following evaluation indicators are defined:

[0092] Conservative forecast rate (CONSERV%):

[0093]

[0094] Forecast conservative error mean (EA_CONSERV):

[0095]

[0096] Maximum conservative error of prediction (EM_CONSERV):

[0097] EM_CONSERV=Max(R_Pred_conserv[n]-R_conserv[n]),n∈[1, m] (9)

[0098] Where num_conserv is the number of conservative samples for stability margin prediction; m is the number of test samples; R_Pred_conserv[n] and R_conserv[n] are the predicted value and true value of the conservative stability margin of the nth test sample, respectively.

[0099] 3) Overall forecast error mean (EA), overall forecast error maximum (EM):

[0100] Overall prediction error mean (Error Average, EA):

[0101]

[0102] The maximum value of the overall prediction error (Error Max, EM):

[0103] EM=Max(R_Pred[n]-R[n]), n∈[1, m] (11)

[0104] Where R_Pred[n] is the predicted value of the stability margin of the nth test sample; R[n] is the true value of the stability margin of the nth test sample.

[0105] It should be noted that the sum of the prediction optimism rate and the prediction conservative rate is 1, among which the larger the prediction conservative rate value, the better, that is, the stability margin prediction result is more in line with the requirements of the power grid for safe and stable operation. In other words, the smaller the prediction optimism rate, the more in line with the requirements of the power grid for safe and stable operation.

[0106] In addition, the smaller the mean of the optimistic prediction error, the mean of the conservative prediction error, and the mean of the overall prediction error, the higher the prediction accuracy; the smaller the maximum value of the optimistic prediction error, the maximum value of the conservative prediction error, and the maximum value of the overall prediction error, the higher the stability of the prediction result.

[0107] In summary, the smaller the error mean of the stability margin prediction value, the higher the prediction accuracy. The smaller the maximum error of the stability margin prediction value, the higher the stability of the prediction result. The higher the prediction conservative rate, the more the stability margin prediction result meets the requirements of the power grid for safe and stable operation.

[0108] In order to verify the effectiveness and superiority of the Edge-GCN-Transformer model in the stability margin prediction method of the present invention, a comparative model without considering the influence of HRPS topology was established: a Transformer-based HRPS small disturbance stability margin prediction model. Figure 3 The site topology shown is similar to the sample set generated. Each short circuit ratio scenario has 2048 samples (2 10 ×2=2048), there are 22528 samples in total for the 11 short circuit ratio scenarios. The samples of each short circuit ratio scenario are divided into training set and test set according to 3:1, that is, samples are extracted from each short circuit ratio scenario sample in a ratio of 1536:512 to form the training set and test set. Similarly, there are 2048 samples for each short circuit ratio scenario in the generalized test set, and there are 4096 samples for the two short circuit ratio scenarios.

[0109] The values ​​of various evaluation indexes of the HRPS small interference stability margin prediction results based on the Edge-GCN-Transformer model and the Transformer model are shown in Tables 3 to 8. Since there are no low-frequency band samples in the test set and the generalized test set, only the stability margin prediction results of the sub / super-synchronous frequency bands are compared and analyzed.

[0110] Table 3 Optimism evaluation index value table of stability margin prediction results (test set)

[0111]

[0112]

[0113] Table 4 Conservative evaluation index values ​​of stability margin prediction results (test set)

[0114]

[0115] Table 5 Stability margin prediction accuracy evaluation index values ​​(test set)

[0116]

[0117] Table 6 Optimism evaluation index values ​​of stability margin prediction results (generalization test set)

[0118]

[0119]

[0120] Table 7 Conservative evaluation index values ​​of stability margin prediction results (generalized test set)

[0121]

[0122] Table 8 Stability margin prediction accuracy evaluation index values ​​(generalization test set)

[0123]

[0124] It can be seen from Tables 3 to 8 that for the sub- / super-frequency strongly damped samples in the test set, the mean optimistic error, conservative error and overall prediction error of the stability margin prediction based on the Edge-GCN-Transformer prediction model are 0.000191, 0.003667 and 0.002915, respectively, which are 0.000355, 0.004949 and 0.003236 based on the Transformer prediction model, and the mean prediction error is reduced by 46.20%, 25.90% and 9.92%, respectively. Taking the sub- / super-frequency weak damping samples in the generalization test set as an example, the stability margin prediction optimism rate and conservative prediction rate based on the Edge-GCN-Transformer prediction model are 36.35% and 63.65% respectively. Compared with 49.67% and 50.33% based on the Transformer prediction model, the prediction optimism rate decreased by 26.82% and the prediction conservative rate increased by 26.47%.

[0125] In general, for the test sample set and the generalized sample set, compared with the Transformer-based prediction model, the indicators of the HRPS small disturbance stability margin prediction model based on Edge-GCN-Transformer have been improved, and the prediction results are more conservative and significantly higher than the optimistic rate, which meets the requirements for safe and stable operation of the power grid. The higher prediction accuracy and prediction conservative rate show the superiority of this method.

[0126] Furthermore, the present invention also provides a hybrid new energy station small disturbance stability margin prediction device. Figure 4As shown, the hybrid new energy station small disturbance stability margin prediction device 100 includes a data processing module 10, a construction module 20, an embedding module 30 and a prediction module 40 which are connected in sequence, wherein the prediction module 40 is also connected to the data processing module 10.

[0127] In this embodiment, the data processing module 10 is used to obtain the system operation status data and perform dimensionality reduction processing on the system operation status data; the construction module 20 is used to construct an edge convolution module based on the edge graph convolutional network; the embedding module 30 is used to embed the edge convolution module into the Transformer model based on the topological enhanced self-attention mechanism to obtain an optimized Transformer model; the prediction module 40 is used to input the system operation status data after dimensionality reduction processing into the optimized Transformer model, and obtain the predicted value of the grid-connected stability margin of the hybrid new energy station through the optimized Transformer model, thereby realizing the small disturbance stability margin prediction of the hybrid new energy station.

[0128] Among them, when constructing an edge convolution module based on an edge graph convolutional network, the construction module 20 is specifically used to: obtain HRPS topology data and collector network flow data; calculate edge weights based on branch flow information in the collector network flow data to assist the edge convolution module to be constructed to learn the HRPS topology data; perform convolution operations on the features of nodes and edges in the HRPS topology data to update the nodes, wherein the nodes are power generation units in the HRPS topology structure, and the edges are lines between the power generation units; use the ReLU activation function to perform nonlinear transformation processing on the updated nodes, and use the processed updated node features as input to perform convolution operations on the next layer. After multiple convolution layer operations, an edge convolution module is formed.

[0129] It should be noted that the specific implementation of the hybrid new energy station small disturbance stability margin prediction device of the embodiment of the present invention can refer to the specific implementation of the hybrid new energy station small disturbance stability margin prediction method mentioned above, which will not be repeated here to avoid redundancy.

[0130] In summary, on the one hand, in order to solve the problem of too high dimensionality of model input feature data, the present invention adopts T-distributed random neighbor embedding algorithm to perform data dimensionality reduction processing, thereby reducing computational complexity and improving the quality of input feature data. On the other hand, the present invention constructs voltage and power holomorphic functions by embedding complex variables in the collector network flow equation, and uses the holomorphic embedding method to realize fast calculation of the flow; on this basis, an edge graph convolutional network is introduced to construct an edge convolution module, which not only considers the topological connection in HRPS, but also takes the flow information calculated by the holomorphic embedding method into consideration as the edge feature of the edge convolution module. Compared with the traditional graph neural network, the edge convolution module based on the edge graph convolutional network not only realizes the perception of HRPS topology, adapts to the topological changes within the station, and improves the topological generalization ability of the model, but also can obtain the flow characteristics of the collector network, providing richer feature input for subsequent small disturbance stability margin prediction. On the basis of the first two, the present invention aims at the self-attention mechanism in the traditional Transformer model, which quantifies the global correlation between input data through dot product calculation and ignores the topological correlation problem of input features. The edge convolution module based on the edge graph convolution network is embedded into the self-attention mechanism of the Transformer model. The nodes and edge information in the graph are introduced into the self-attention calculation by constructing an edge convolution module and improving the attention matrix, so that the optimized Transformer model can better fit the nonlinear relationship between the edge graph characteristics of HRPS and the operating status characteristics of the grid-connected system and the small disturbance stability margin, thereby improving the prediction accuracy of the Edge-GCN-Transformer deep learning model.

[0131] Furthermore, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned method for predicting small disturbance stability margin of a hybrid new energy station is implemented.

[0132] Furthermore, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for predicting small disturbance stability margin of a hybrid new energy station is implemented.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0134] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A method for predicting small disturbance stability margin of hybrid new energy station, characterized in that: include: Obtain system operation status data and perform dimensionality reduction processing on the system operation status data; Construct an edge convolution module based on edge graph convolutional network; The edge convolution module is embedded into the Transformer model based on the topology-enhanced self-attention mechanism to obtain the optimized Transformer model; The system operation status data after dimensionality reduction processing is input into the optimized Transformer model. The predicted value of the grid-connected stability margin of the hybrid new energy station is predicted by the optimized Transformer model, thereby realizing the small disturbance stability margin prediction of the hybrid new energy station.

2. The method for predicting small disturbance stability margin of a hybrid new energy station according to claim 1, characterized in that: The system operation status data includes the strength of the external power grid, the control type of the generator sets in the HRPS, the capacity ratio and location distribution, and the control type of the energy storage device at the grid connection point.

3. The method for predicting small disturbance stability margin of a hybrid new energy station according to claim 1, characterized in that: The system operation status data is reduced in dimension based on the T-distributed random neighbor embedding algorithm.

4. The method for predicting small disturbance stability margin of a hybrid new energy station according to claim 1, characterized in that: Construct an edge convolution module based on edge graph convolutional network, including: Obtain HRPS topology data and collector network flow data; The edge weight is calculated according to the branch flow information in the collector network flow data to assist the edge convolution module to be constructed to learn the HRPS topology data; Convolution operation is performed on the features of nodes and edges in the HRPS topology data to update the nodes, wherein the nodes are power generation units in the HRPS topology structure and the edges are lines between the power generation units; The ReLU activation function is used to perform nonlinear transformation processing on the update node, and the processed updated node features are used as input to perform the convolution operation of the next layer. After multiple layers of convolution operations, the edge convolution module is formed.

5. The method for predicting small disturbance stability margin of a hybrid new energy station as claimed in claim 4, characterized in that: Based on the holomorphic embedding method, power flow calculation is performed according to the HRPS topology data to obtain the power collection network power flow data.

6. The method for predicting small disturbance stability margin of a hybrid new energy station according to claim 1, characterized in that: The optimized Transformer model weights the attention score through the embedded side convolution module when performing self-attention calculation.

7. A hybrid new energy station small disturbance stability margin prediction device, characterized in that: include: A data processing module is used to obtain system operation status data and perform dimensionality reduction processing on the system operation status data; A building module for building edge convolution modules based on edge graph convolutional networks; The embedding module is used to embed the edge convolution module into the Transformer model based on the topological enhanced self-attention mechanism to obtain the optimized Transformer model; The prediction module is used to input the system operation status data after dimensionality reduction processing into the optimized Transformer model, and obtain the predicted value of the grid-connected stability margin of the hybrid new energy station through the optimized Transformer model, thereby realizing the small disturbance stability margin prediction of the hybrid new energy station.

8. The hybrid new energy station small disturbance stability margin prediction device according to claim 7, characterized in that: When constructing an edge convolution module based on an edge graph convolutional network, the construction module is specifically used to: Obtain HRPS topology data and collector network flow data; The edge weight is calculated according to the branch flow information in the collector network flow data to assist the edge convolution module to be constructed to learn the HRPS topology data; Convolution operation is performed on the features of nodes and edges in the HRPS topology data to update the nodes, wherein the nodes are power generation units in the HRPS topology structure and the edges are lines between the power generation units; The ReLU activation function is used to perform nonlinear transformation on the update node, and the processed updated node features are used as input for the next layer of convolution operation. After multiple layers of convolution operations, an edge convolution module is formed.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting small disturbance stability margin of a hybrid new energy station as described in any one of claims 1 to 6 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for predicting small disturbance stability margin of a hybrid new energy station as described in any one of claims 1-6 is implemented.

Images